A Multi-Scale Residual Neural Network for ECG Based Person Identification

The electrocardiogram (ECG) signal is unique for each person. Due to the physiological and geometrical differences of the heart, the morphological shape of the ECG signal changes for different persons. In this work, we have proposed a multiscale residual neural network (MS-ResNet) architecture to exploit the morphological shape of the ECG waveforms and their inter-relationship for biometric application. The proposed neural network model utilises three parallel convolutional branches with dilated kernels for capturing the multi-scale representation of the ECG signal. The proposed MS-ResNet model is evaluated using two publicly available ECG datasets, i.e., the UofTDB dataset and CYBHi dataset. The proposed model gives an identification accuracy of 87.98% and 94.18% for the UofTDB and CYBHi datasets, respectively.